Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75977.88891384043,
-      "required_pages_per_second_upper_estimate": 0.0052563266480501955
+      "required_token_positions_per_second": 75980.60839700211,
+      "required_pages_per_second_upper_estimate": 0.005256514788205391
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9497.236114230054,
-      "required_pages_per_second_upper_estimate": 0.003976294436185789
+      "required_token_positions_per_second": 9497.576049625264,
+      "required_pages_per_second_upper_estimate": 0.003976436760037105
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18994.47222846011,
-      "required_pages_per_second_upper_estimate": 0.003956047647270305
+      "required_token_positions_per_second": 18995.15209925053,
+      "required_pages_per_second_upper_estimate": 0.003956189246426551
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28491.708342690163,
-      "required_pages_per_second_upper_estimate": 0.0017080344108246178
+      "required_token_positions_per_second": 28492.72814887579,
+      "required_pages_per_second_upper_estimate": 0.0017080955466482924
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28491.708342690163,
-      "required_pages_per_second_upper_estimate": 0.0022151050018635746
+      "required_token_positions_per_second": 28492.72814887579,
+      "required_pages_per_second_upper_estimate": 0.002215184287308854
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28491.708342690163,
-      "required_pages_per_second_upper_estimate": 0.36546104260707485
+      "required_token_positions_per_second": 28492.72814887579,
+      "required_pages_per_second_upper_estimate": 0.3654741235858415
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09280191896110068,
-    "producer_p95_seconds": 0.22937024799830397,
-    "gpu_update_seconds": 0.9703875832034261,
-    "producer_batches_per_second": 10.775639245338935,
-    "gpu_batches_per_second": 1.0305160714225319,
-    "producer_headroom": 10.45654652475644,
+    "producer_batch_seconds": 0.23307675319540522,
+    "producer_p95_seconds": 0.2512632610014407,
+    "gpu_update_seconds": 0.9703528512797354,
+    "producer_batches_per_second": 4.29043217004841,
+    "gpu_batches_per_second": 1.0305529567735747,
+    "producer_headroom": 4.163233089428776,
     "sustainable": true,
-    "buffer_seconds": 60.16403015861242,
+    "buffer_seconds": 60.161876779343594,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418274.0892868912,
-    "upload_bytes_per_second": 3448446.3074146025,
+    "generation_bytes_per_second": 415907.04725756036,
+    "upload_bytes_per_second": 3428564.930559644,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.75188263850262,
+    "estimated_full_shards_per_hour": 89.24397052092655,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.166742791001525,
-      "bytes": 5284765,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.747680902999491,
       "bytes": 7697898,
@@ -562,6 +556,12 @@
       "bytes": 10549433,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.0944702749984572,
+      "bytes": 2560030,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -167,3 +167,4 @@
 {"step": 859200, "loss": 0.5276119112968445, "lr": 3e-05, "grad_norm": 0.30946591336955503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125762791077, "tokens_per_second": 150743.05116916314, "data_wait_s": 1.3836997823091224e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 859300, "loss": 0.5622125864028931, "lr": 3e-05, "grad_norm": 0.3176077509608985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125777440206, "tokens_per_second": 152282.01830021347, "data_wait_s": 1.2996999430470169e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 859400, "loss": 0.521033763885498, "lr": 3e-05, "grad_norm": 0.29053072749407305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125792029739, "tokens_per_second": 149965.38916220784, "data_wait_s": 1.9274997612228617e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 859500, "loss": 0.4725848436355591, "lr": 3e-05, "grad_norm": 0.2987977973834179, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125806659355, "tokens_per_second": 152020.10628561885, "data_wait_s": 1.4536999515257776e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:676827b9ede6ae73bd4e916a36d2e5e38b2c89c80564b927691229da2e45f2ef
-size 360456
+oid sha256:ffaa900765db036511e6bf1203a4f54ee86fdf9dbf556e3f12241ac247f228f8
+size 362574